主讲人介绍：Prof. Jiying Zhao received his B.Eng. and M.Eng. in Computer Engineering, and Ph.D. degree in Electrical Engineering from North China Electric Power University (NCEPU), China, respectively in 1983, 1988, and 1992, and his second Ph.D. degree in Engineering from Keio University, Japan, in 1998. He worked for NCEPU as an assistant lecturer, lecturer, and associate professor from 1983 to 1994, and as an instructor and associate professor for Department of Computing Science at University College of the Cariboo (now Thompson Rivers University), Canada, from 1998 to 2001. He joined the School of Electrical Engineering and Computer Science (EECS) at University of Ottawa as an assistant professor in 2001, became an associate professor in 2005, and then a full professor in 2009. His current research interests are on image/video processing and multimedia communications.
内容摘要：In this presentation, we report a chroma keying system that automatically estimates the alpha map and the reliable intrinsic color of foreground objects in front of solid background [1]. Our system is designed to be capable of distinguishing the transparent foreground from the reflective foreground and shaded background, thereby making the artifacts of the composited image less conspicuous. Specifically, we assume that the transparent region tends to be with higher saturation and lightness compared with region reflecting background light. With this assumption, a threshold function (TF) on a saturation-lightness plane is defined according to human visual experiments. The pixels with color mixed with the background light (conventional unknown pixels) are now further categorized into reflective pixels and transparent pixels according to TF. In this case, the reflective and the transparent regions are separated to improve the alpha matte quality. According to our simulation, the proposed chroma keying system generates high quality composited images that are little affected by reflecting, background shading, and intrinsic color missing. At the end, we are going to report our recent progress on deep learning based image keying.
[1] Wenyi Wang and Jiying Zhao, Robust Image Chroma-Keying: A Quadmap Approach Based on Global Sampling and Local Affinity, IEEE Transactions on Broadcasting, Vol.61, Iss.3, pp. 356-366, September 2015.
